RE

repo-stage-review-loop

Provides a formal review process for plans and implementation code, ensuring changes meet the agreed contract.

Install

mkdir -p .claude/skills/repo-stage-review-loop && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/13572" && unzip -o skill.zip -d .claude/skills/repo-stage-review-loop && rm skill.zip

Installs to .claude/skills/repo-stage-review-loop

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Use when a RepoPilot plan or implementation needs formal review, when final code changed after earlier review, or when external findings require evidence-based triage.
167 chars✓ has a “when” trigger
Advanced

Key capabilities

  • Review the plan contract before implementation.
  • Review the final implementation state against the approved contract after code changes.
  • Report severity-ordered findings with file/line evidence.
  • Classify external findings as fix, clarify, reject, or defer.
  • Perform a focused Stage Debt Sweep over changed paths.

How it works

This skill systematically reviews plans or code against an approved contract, identifying issues by severity and providing evidence, then triages external feedback.

Inputs & outputs

You give it
Proposed plan, active OpenSpec artifacts, changed files, or external findings.
You get back
Severity-ordered findings with evidence, or a classification of external findings.

When to use repo-stage-review-loop

  • Review proposed architecture plans
  • Validate final code changes against specs
  • Triage external feedback into fix/clarify/reject

About this skill

Repo Stage Review Loop

Core Rule

Review the plan contract before implementation when requested, and review the final implementation state against the approved contract after code changes. Passing tests and completed tasks are inputs to review, not substitutes for it.

Plan Contract Review

For plan review, read the proposed plan or active OpenSpec artifacts, Harness boundaries, relevant specs, and directly implicated runtime/docs. Report severity-ordered findings against intent, scope, non-goals, test plan, review gates, and roadmap truth. Medium/high plans should receive internal, Codex independent, and OpenCode independent review before implementation.

Review Loop

  1. Read the active OpenSpec contract, changed files or plan contract, tests, allowed paths, and current review checklist.
  2. Confirm the review occurs after the latest runtime/test change.
  3. Review in layers: scope, business logic, architecture boundary, minimality, failure semantics, security/privacy, test adequacy, and maintainability. For user-facing summaries in Chinese, keep precise English terms and add a short Chinese explanation or concrete example when the term is non-obvious.
  4. Report severity-ordered findings with file/line evidence, trigger, consequence, and missing regression coverage. If there are no findings, state inspected areas and residual risk.
  5. Use external-review-triage for external findings. Classify each as fix, clarify, reject, or defer; never accept it by authority alone.
  6. After remediation, rerun affected verification and review changed behavior.
  7. Perform a focused Stage Debt Sweep over changed paths and directly dependent older paths. Record inspected paths, concrete findings, dispositions, and residual debt.
  8. Block archive when tasks are unchecked, review evidence is stale, validation failed, blocking findings remain, or delta operations do not match long-term specs.

Review Priorities

  • stage scope and user-visible behavior match the approved contract
  • business logic, state transitions, and failure paths match intended semantics
  • code remains in the correct architectural layer and reuses existing boundaries
  • functions/classes stay minimal enough to avoid hidden behavior coupling
  • public contract and state-transition correctness
  • fail-closed permissions, approval, identity, path, and lifecycle checks
  • interruption, retry, rollback, and reconciliation behavior
  • tests that assert the intended contract rather than implementation details
  • scope drift and accidental roadmap capability claims
  • stale assumptions in directly dependent older paths

External review should seek independent counterexamples, especially for medium/high-risk stages. Repeating the task checklist is not useful diversity.

Evidence Boundary

Store gate evidence in .harness/review_checklist.md. Store durable unresolved debt in docs/PROGRESS.md. Put only next-session blockers in HANDOFF_TO_NEXT_CHAT.md.

Do not perform merge/push handoff here. Return to repo-stage-workflow, which uses repo-stage-handoff after integration.

Evals

Use references/evals.md when changing routing or review gates.

When not to use it

  • When the task is not related to formal review of plans or code implementation.
  • When the task is to perform merge/push handoff.

Limitations

  • The skill does not perform merge/push handoff.
  • The skill requires the review to occur after the latest runtime/test change.

How it compares

This skill enforces a multi-layered review process with severity-ordered findings and external feedback triage, unlike a simple checklist review.

Compared to similar skills

repo-stage-review-loop side by side with the closest alternatives in the catalog.

SkillInstallsUpdatedSafetyDifficulty
repo-stage-review-loop (this skill)01moNo flagsAdvanced
test-plan16moNo flagsIntermediate
flow-next-prime02moReviewIntermediate
gate-review05moNo flagsBeginner

Try saying

Example prompts that trigger this skill in your AI assistant.

Search skills

Search the agent skills registry